Author Archive

You can add there specific instructions about your website’s focus, target audience, or special features to help the AI provide more relevant and personalized answers to your visitors.

Example:

The real reason ChatGPT feels instant is streaming – words appear as generated. We use GPT via API and script needs to wait for the complete response. Also keep in mind that searching/fetching listing data needs to be processed in the background – that adds latency aswell.

The AI assistant can filter listings using these parameters when calling the search function:

– location 
– rating 
– price_min / price_max – Price range filtering
– date_start / date_end – Availability dates or event dates

The AI automatically understands user intent and applies relevant filters working in tandem with semantic search. For example, asking “find restaurants in New York rated 4.5+” uses semantic understanding to identify “restaurants” from description, while simultaneously applying location and rating filter when searching through database.

AI search understands meaning, so it finds relevant results even when the exact words aren’t in your listings. That’s why it can match “romantic dinner spot” with fine dining restaurants or “dog grooming” with pet salons.

But it’s not magic — it works from the text in your listings. If a listing has almost no description, or uses very unusual wording, the AI may miss it. Also, in rare cases, it might return a result that feels related but isn’t exactly what you meant.

In short: it’s far better than keyword search, but like any search, it depends on good, descriptive content.

Currently – NOT but you can use Custom System Prompt feature to provide relevant info about your website to AI model.

Follow this checklist:

1. ✅  Make sure your your listings have AI embeddings generated
2. ✅ Make sure your listings have keyword-rich descriptions!
3. ✅ Try “Query Expansion”  feature
4. ✅ Lower  “Minimum Match Percentage” setting

AI search understands meaning, so it finds relevant results even when the exact words aren’t in your listings. That’s why it can match “romantic dinner spot” with fine dining restaurants or “dog grooming” with pet salons.

But it’s not magic — it works from the text in your listings. If a listing has almost no description, or uses very unusual wording, the AI may miss it. Also, in rare cases, it might return a result that feels related but isn’t exactly what you meant.

In short: it’s far better than keyword search, but like any search, it depends on good, descriptive content.

This setting determines how closely a listing must match the search query to appear in
results.

Setting Range: 0-100 (Default: 65)
Impact on Search Results:
– Lower Values (40-60): More results, some may be less relevant
– Medium Values (65-75): Balanced relevance and quantity (recommended)
– Higher Values (80-95): Fewer but highly relevant results

Example:
– Query: “Italian restaurant”
– 50% threshold: Returns pizza places, pasta shops, Mediterranean food
– 80% threshold: Only returns specifically Italian restaurants

It’s an experimental feature. Keep in mind that it works on a text basis, not coordinates. The AI extracts the location from the search query and then matches it as a keyword to listing location – it’s less accurate than “normal” location search with Google API + Radius.

Impact on Search Results:
– ✅ Enabled: “Pizza in downtown” automatically filters by downtown area
– ❌ Disabled: “downtown” is treated as just another keyword

Query expansion uses AI to understand and broaden search terms for better results.

Example:
– Original query: “coffee shop”
– Expanded query might include: “café”, “espresso bar”, “coffee house”

Trade-offs:
– ✅ Enabled: Better coverage, finds more results but not necessary always relevant (depends on minimal match settings)
– ❌ Disabled: Faster processing, more literal searches

After installing plugin, adding Open AI api key and clicking “Start Generation” the plugin sends structured listing data to OpenAI’s text-embedding-3-small model API to create vector embeddings: listing name, category, location,  amenities, features, description, FAQ, pricing/menu, user reviews (positive only, rating ≥3.0)

When this happens:
– Initial setup when enabling AI search
– Manual regeneration via admin panel
– Daily cron job (listeo_ai_update_embeddings) for changed listings
– When a listing is created/updated (if auto-update enabled)

What happens when user is searching?
– Listings data is not sent again when user searches for something
– User’s search query text is sent to OpenAI to generate an embedding vector
– This vector is compared locally against stored embeddings (no additional API calls)